namespace compression

// namespaces

namespace xf::compression::details

// typedefs

typedef ap_uint <8> streamDt
typedef ap_uint <c_frequency_bits> Frequency
typedef ap_uint <12> Histogram
typedef ap_uint <32> compressd_dt
typedef struct xf::compression::lz4BlockInfo dt_lz4BlockInfo
typedef ap_uint <kGMemDWidth> uintMemWidth_t
typedef struct xf::compression::BlockInfo dt_blockInfo

// enums

enum eHuffmanType

// structs

struct BlockInfo
struct Codeword
struct Symbol
struct lz4BlockInfo

// global variables

static const uint8_t c_tgnSymbolBits
static const uint8_t c_tgnBitlengthBits
static const uint16_t c_tgnSymbolSize
static const uint16_t c_tgnTreeDepth
static const uint16_t c_tgnMaxBits
const auto BSIZE_STD_64KB
const auto BSIZE_STD_256KB
const auto BSIZE_STD_1024KB
const auto BSIZE_STD_4096KB
const auto MAX_BSIZE_64KB
const auto MAX_BSIZE_256KB
const auto MAX_BSIZE_1024KB
const auto MAX_BSIZE_4096KB
const auto MAGIC_HEADER_SIZE
const auto MAGIC_BYTE_1
const auto MAGIC_BYTE_2
const auto MAGIC_BYTE_3
const auto MAGIC_BYTE_4
const auto FLG_BYTE
const int NO_COMPRESS_BIT
const auto BSIZE_NCOMP_64
const auto BSIZE_NCOMP_256
const auto BSIZE_NCOMP_1024
const auto BSIZE_NCOMP_4096
const int kGMemDWidth
const int c_lTreeSize
const int c_dTreeSize
const int c_bLTreeSize
const int c_maxCodeSize

Global Variables

const auto BSIZE_STD_64KB

Below are the codes as per LZ4 standard for various maximum block sizes supported.

const auto MAX_BSIZE_64KB

Maximum block sizes supported by LZ4

const int NO_COMPRESS_BIT

This value is used to set uncompressed block size value. 4th byte is always set to below and placed as uncompressed byte

const auto BSIZE_NCOMP_64

In case of uncompressed block Values below are used to set 3rd byte to following values w.r.t various maximum block sizes supported by standard

blockPacker

#include "block_packer.hpp"

template <int DATAWIDTH = 512>
void blockPacker (
    hls::stream <ap_uint <DATAWIDTH>>& inStream,
    hls::stream <uint32_t>& inStreamSize,
    hls::stream <ap_uint <DATAWIDTH>>& outStream,
    hls::stream <bool>& outStreamEos,
    hls::stream <uint32_t>& outCompressedSize
    )

Compression Packer module packs the compressed data.

Parameters:

DATAWIDTH	input data width
inStream	input data
outStream	output data
inStreamSize	size of the data in input stream
outStreamEos	output end of stream
outCompressedSize	total compressed packed data size

huffmanDecoder

#include "huffman_decoder.hpp"

template <
    eHuffmanType DECODER = FULL,
    int ByteGenLoopII = 2,
    bool USE_GZIP = 0
    >
void huffmanDecoder (
    hls::stream <ap_uint <16>>& inStream,
    hls::stream <compressd_dt>& outStream,
    hls::stream <bool>& endOfStream,
    uint32_t input_size
    )

This module is ZLIB/GZIP Fixed, Dynamic and Stored block supported decoder. It takes ZLIB/GZIP Huffman encoded data as input and generates decoded data in LZ77 format (Literal, Length, Offset).

Parameters:

DECODER	Fixed, Full, Dynamic huffman block support
ByteGenLoopII	core bytegenerator loop initiation interval
USE_GZIP	switch that controls GZIP/ZLIB header processing
inStream	input bit packed data
outStream	output lz77 compressed output in the form of 32bit packets (Literals, Match Length, Distances)
endOfStream	output completion of execution
input_size	input data size

huffmanEncoder

#include "huffman_encoder.hpp"

void huffmanEncoder (
    hls::stream <encodedV_dt>& inStream,
    hls::stream <uint16_t>& outStream,
    hls::stream <uint8_t>& outStreamSize,
    uint32_t input_size,
    hls::stream <uint16_t>& inStreamCodes,
    hls::stream <uint8_t>& inStreamCodeSize
    )

This module does zlib/gzip dynamic huffman encoding.

Parameters:

inStream	input packet of 32bit size which contains either literal or match length and distance information. Example: [Literal (1 Byte) | ML (1 Byte) | DIST (2 Bytes)]
outStream	output bit encoded LZ77 compressed data
outStreamSize	output Stream Size
input_size	uncompressed data input size
inStreamCodes	Huffman Codes
inStreamCodeSize	HuffmanCode Lengths

lz4Compress

#include "lz4_compress.hpp"

template <
    int MAX_LIT_COUNT,
    int PARALLEL_UNITS
    >
static void lz4Compress (
    hls::stream <compressd_dt>& inStream,
    hls::stream <ap_uint <8>>& outStream,
    uint32_t max_lit_limit [PARALLEL_UNITS],
    uint32_t input_size,
    hls::stream <bool>& endOfStream,
    hls::stream <uint32_t>& compressdSizeStream,
    uint32_t index
    )

This is the core compression module which seperates the input stream into two output streams, one literal stream and other offset stream, then lz4 encoding is done.

Parameters:

inStream	Input data stream
outStream	Output data stream
max_lit_limit	Size for compressed stream
input_size	Size of input data
endOfStream	Stream indicating that all data is processed or not
compressdSizeStream	Gives the compressed size for each 64K block

lz4Decompress

#include "lz4_decompress.hpp"

void lz4Decompress (
    hls::stream <ap_uint <8>>& inStream,
    hls::stream <compressd_dt>& outStream,
    uint32_t input_size
    )

This module reads the compressed data from input stream and decodes the offset, match length and literals by processing in various decompress states.

Parameters:

inStream	Input stream 8bit
outStream	Output stream 32bit
input_size	Input size

lzDecompress

#include "lz_decompress.hpp"

template <
    int HISTORY_SIZE,
    int LOW_OFFSET = 8
    >
void lzDecompress (
    hls::stream <compressd_dt>& inStream,
    hls::stream <ap_uint <8>>& outStream,
    uint32_t original_size
    )

This module writes the literals to the output stream as it is and when match length and offset are read, the literals will be read from the local dictionary based on offset until match length.

Parameters:

LOW_OFFSET	low offset
HISTORY_SIZE	history size
inStream	input stream
outStream	output stream
original_size	original size

lzMultiByteDecompress

#include "lz_decompress.hpp"

template <
    int PARALLEL_BYTES,
    int HISTORY_SIZE,
    class SIZE_DT = uint8_t
    >
void lzMultiByteDecompress (
    hls::stream <SIZE_DT>& litlenStream,
    hls::stream <ap_uint <PARALLEL_BYTES*8>>& litStream,
    hls::stream <ap_uint <16>>& offsetStream,
    hls::stream <SIZE_DT>& matchlenStream,
    hls::stream <ap_uint <PARALLEL_BYTES*8>>& outStream,
    hls::stream <bool>& endOfStream,
    hls::stream <uint64_t>& sizeOutStream
    )

This module writes the literals to the output stream as it is and when match length and offset are read, the literals will be read from the local dictionary based on offset until match length. This module can process data in parallel defined by PARALLEL_BYTES template argument.

Parameters:

PARALLEL_BYTES	number of bytes processed in parallel (4, 8)
HISTORY_SIZE	history size
SIZE_DT	input data type
litlenStream	literal length stream
litStream	literals only stream
offsetStream	offset only stream
matchlenStream	match length only stream
outStream	output stream
endOfStream	end of stream
sizeOutStream	output size stream

lzBestMatchFilter

#include "lz_optional.hpp"

template <
    int MATCH_LEN,
    int OFFSET_WINDOW
    >
void lzBestMatchFilter (
    hls::stream <compressd_dt>& inStream,
    hls::stream <compressd_dt>& outStream,
    uint32_t input_size
    )

Objective of this module is to pick character with higher match length in the offset window range.

Parameters:

inStream	input stream
outStream	output stream
input_size	input stream size

lzBooster

#include "lz_optional.hpp"

template <
    int MAX_MATCH_LEN,
    int BOOSTER_OFFSET_WINDOW = 16 * 1024,
    int LEFT_BYTES = 64
    >
void lzBooster (
    hls::stream <compressd_dt>& inStream,
    hls::stream <compressd_dt>& outStream,
    uint32_t input_size
    )

This module helps in improving the compression ratio. Finds a better match length by performing more character matches with supported max match, while maintaining an offset window. Booster offset Window template argument (default value is 16K) internally consume BRAM memory to implement history window. Higher the booster value can give better compression ratio but will consume more BRAM resources.

Parameters:

MAX_MATCH_LEN	maximum length allowed for character match
BOOSTER_OFFSET_WINDOW	offset window to store/match the character
inStream	input stream 32bit per read
outStream	output stream 32bit per write
input_size	input size
left_bytes	last 64 left over bytes

lzFilter

#include "lz_optional.hpp"

template <int LEFT_BYTES = 64>
static void lzFilter (
    hls::stream <compressd_dt>& inStream,
    hls::stream <compressd_dt>& outStream,
    uint32_t input_size
    )

This module checks if match length exists, and if match length exists it filters the match length -1 characters writing to output stream.

Parameters:

MATCH_LEN	length of matched segment
OFFSET_WINDOW	output window
inStream	input stream
outStream	output stream
input_size	input stream size
left_bytes	bytes left in block

snappyCompress

#include "snappy_compress.hpp"

template <
    int MAX_LIT_COUNT,
    int MAX_LIT_STREAM_SIZE,
    int PARALLEL_UNITS
    >
static void snappyCompress (
    hls::stream <compressd_dt>& inStream,
    hls::stream <ap_uint <8>>& outStream,
    uint32_t max_lit_limit [PARALLEL_UNITS],
    uint32_t input_size,
    hls::stream <bool>& endOfStream,
    hls::stream <uint32_t>& compressdSizeStream,
    uint32_t index
    )

This is the core compression module which seperates the input stream into two output streams, one literal stream and other offset stream, then encoding is done based on the snappy algorithm.

Parameters:

inStream	Input data stream
outStream	Output data stream
max_lit_limit	Size for compressed stream
input_size	Size of input data
endOfStream	Stream indicating that all data is processed or not
compressdSizeStream	Gives the compressed size for each 64K block

snappyDecompress

#include "snappy_decompress.hpp"

static void snappyDecompress (
    hls::stream <uintV_t>& inStream,
    hls::stream <encoded_dt>& outStream,
    uint32_t input_size
    )

This module decodes the compressed data based on the snappy decompression format.

Parameters:

inStream	input stream
outStream	output stream
input_size	input data size